Intersample Clipping - A Problem with Most Digital Playback Systems?

[@Audioholics]

Please file a Petition to the FTC by November 8th, 2024 if this topic concerns you. File the petition here: www.regulations.gov/commenton/FTC-2024-0039-0001 "Please reopen the 2024 FTC Final Amplifier Rule (16 CFR Part 432) to address the technical issues raised by Audioholics.com and the CTA, aiming to more accurately rate amplifier power and the maximum SPL performance of powered loudspeaker systems."

[@David_vdk]

is that a yamaha in the back??

[@peterandersson6762]

Hi Gene! If you arnt registerd at Arendals forum, pleas get members there! Taket your old and new friend rewiwers with you. PLEASE🙏

[@onemic-theminimalist]

As a mastering engineer for more than twenty years, I never produce digital masters above -1.0 dB BELOW full scale (0). That's a general practice among professional mastering engineers, although some may go a bit hotter. Modern mastering limiter plug-ins compensate for or warn the engineer of intersample (True Peak) issues. FYI - Allowing headroom and using intersample detection for clean downstream playback has been going on for many years.

[@gurratell7326]

Yeah it's so weird that people for some unclear reason HAVE to have so many samples up to 0dB when they have over 90dB to work with. -1dB is mostly just fine, but you can go down to -3dB to be even safer and without really sacrificing any noise performance.

[@TheGlotz69]

John is still stating that hardware results from an additional 1-3dB of errors without considerations in the studio.

[@onemic-theminimalist]

@@TheGlotz69 I'm not sure I understand your comment.

[@Raypirri]

This is fabulous Gene. Thanks too go to John for spending the time and patience in demonstrating this. Love your work, Dude. Cheers.

[@Malamba4231]

I really enjoy these tech talks with John Siau. Please keep them coming, thank you.

[@AlexandreLollini]

If only the whole world and sciences could have this simple but pure reasoning, in testing health or food products, based on measurements.

[@ericmc6482]

Reaper Digital Audio Workstation software has selectable Peak/True Peak metering and report of track TP Overs. Examining exisiting tracks is very informative showing the extent of overloads and goes long way to explaining bad/harsh sound in so many recordings.....and also reinforces nice/good sounding recordings having sensible mastering levels. The loudness wars exploited DAW peak Normalising to -0.1dB peaks without consideration of True Peak overloading......the result was the awful sound we all grew to hate.

[@itsallgood78735]

Please test the Wiim streamer and it's pre-gain setting. Does lowering the pre-gain of the input source (ie, wifi) by 3db happen pre-upsampling and DSP thereby solving this issue? Or, is the pre-gain changing the volume via DSP? There is also a setting called 'volume limit' that is said to reduce volume of the digital signal before PEQ processing that can prevent clipping due to EQ. Does that solve for this? fwiw - I am using the wiim as a streamer w/ DSP -> Digital out to a miniDSP Flex -> digital out to external dacs (Geshelli J3 - full range, Schiit Modi Multibit - Subs). Ideally, I can have the wiim lower the digital signal 3db to minimize the potential intersample distortion in the minidsp or dacs downsteam.

[@scottivlow9962]

I fell back to sleep this morning after watching just the first few minutes. Later tonight I have to rewatch it.

[@grizzly6699]

Ahh!! That explains it!! I've a cheap internet radio from the brand Majority. When using the RCA outputs it would sometimes have a VERY audible distortion. It goes away when I turn the volume on the unit down several steps. It never does this when using the digital out through my Topping DAC and HPA stack. I also turn the DAC out to -3db, as recommended by Gene several years ago. Now I understand the reasoning behind it. Cheers!!

[@wolfmanjacksaid]

Is this a problem for R-2R ladder DACs or just delta sigmas?

[@shodan6401]

However, we have agreed that this issue is indeed present on some recordings. Specifically related to Remastered recordings. So it's not always exclusive to the hardware, just mostly.

[@robjordan63]

Fascinating, well explained, thanks.

[@AlexandreLollini]

The sound of this defect is like crackling in a fireplace, but only happens in loud moments, usually in the last third of a song/tune. it can be very subtle or very obvious. or like static electricity clics, or specifically the kind of sound you get when you stetoscope an internal combustion engine and that there is pinging or pre-ignition.

[@gago-ey9bo]

In the Apple TV settings for the music App, you can turn on sound check. It will curtail the level and equalize levels from any recording, so no overshoots above 0 dbfs happen. This is my understanding, someone correct me if I’m wrong

[@blarky]

Nice topic and presentation, thank you!

[@AlexandreLollini]

I did not know that but I felt it and solved it not knowing what I did. While I don't have the proper dac, This problem can be infact very obvious (not subtle) take M83 Outro, or Celine Dion All by Myself (critical note) using Audacity from the 44.1 cd : lower the level by 3db, try to normalize, it ups back by 3db. But lower by 3db and upsample it to anything higher ant then try to normalize : it will normalize by less than 3db. The same when wanting to take a recording and equalizing BY LOWERING some frequency range : when you normalise back up you find that the levels were somehow increased by the equalizer even if the levels were lowered. The critical misunderstanding of audiophile people is all about the sample, a sample is NOT a pixel, it is a passing point for an analogue signal, in or out, we'll always need analog. Even down there at budget level, you can sacrifice 3db. BUT FIRST up the bits depth (24bits) THEN loose the 3 db, then upsample. SOLVED. I typically listen 85 to 95 db max and CD quality 16bits/44.1khz is like 112db noise floor. I'll continue to do that when I hear the problem on a track. (until I have a proper DAC) And, by the way 44/16bits is enough to preserve music, the most important is to reproduce it well in room, and to have timings perfect, disances from walls etc. Those influence experience much more than sampling and bits. The only value in oversampling and adding bits is to do some work on the signal like filtering, room correction, etc. But for storage, there is no point in exceeding CD. And no point in having more tracks than left and right : we get the complete 3D with 2.1. not yet found anything multichannel beating 2.1 or adding anything of value or interest.

[@Bob.martens]

More! Longer! Deeper! Into the knowledge-hole.

[@shodan6401]

Yeah, 11k is pretty high up, but it's also where most of the inner/micro detail is. I remember when the first oversampling DACs came out, because CDs were so brittle. There was no lot of resentment that the industry standard for CDs was 16 bit, bc execs didn't care about sound quality.

[@tedkay5272]

What about an R2R DAC?

[@Douglas_Blake_579]

Daymn ... ya learns somethin' new every day! Thanks Gene and John for such a detailed and informative presentation. I do, however, have a question... I have a crap ton of digital files on my system, thousands of them... Now the older ones are MP3/320 and most of the newer ones are FLAC/48khz. 1) I know that Windows internally processes these decoders in 32bitFloat format. 2) I also know that the packets of samples from the decoders contain a digital volume setting, which I understand is an offset from the decoded sample values. If I understand this correctly a Dvol of -32 simply subtracts 32 from every sample as it is decoded. There is apparently an "Alignment Level" at which both 0 vu analog and -16dbfs digital should produce a 1 volt rms output level. (Where digital meets analog) I understand that one of the reasons so many digital recordings have embarrassingly high distortion is that they are exceeding 0dbfs before audio conversion. Now I know why. So here is my question... If I use software like MP3GAIN to manipulate the DVol values in these decoder frames to normalize the file to -16dbfs (alignment level) does this avoid the clipping phenomenon you were talking about? A second question is if mastering at the alignment level will also avoid the problem?

[@joscallinet6260]

I am VERY interested to find out if the PCM intersample-overs problem occurs in R-to-R resistor-ladder DACs in the same way as it does in Sigma-Delta-chip-based DACS like the two Benchmark DACs discussed here?

[@BenchmarkMediaSystems]

This is only a problem with oversampled DACs and with any device that upsamples or resamples the audio. In a non-oversampled DAC, the analog lowpass filter reconstructs the inter-sample peaks and this is usually done without clipping (unless the analog output stage clips). If the analog output stage clips, the non-oversampled DAC will just produce harmonic distortion. In contrast, resampling devices will create IMD distortion across the entire audio band. Very different!

[@wolfmanjacksaid]

So if my DAC has a NOS mode, will that result in less or more intersample clipping?

[@ScottGrammer]

This is all curable by just turning the levels down a touch before digitization. That said, when intersample clipping occurs, we're talking about clipping that lasts only for as long as the time between samples - usually a maximum of 1/44,100 of one second. So, the distortion created by the clipping will be of such high frequency that it will not make it past the reconstruction filter in the DAC, and so it never makes it to the analog output of the DAC. But, still, it's easily avoided by just not using up that last dB or so of record level.

[@andersannerstedt5168]

@@ScottGrammer Each time yes, but if it also reoccurs at 80% of the time because the samples missed those peaks as explained you would also have it show up / heard millions of times.

[@hawkins55]

I was using Roon to convert some DSD music to higher DSD rate, with headroom management turned on with 3db headroom. It was clipping like crazy. I had to add another 3db headroom to avoid clipping. The overall volume level was quite a bit lower, but I can turn up the volume. With PCM conversion, I never had issue with 3db headroom added.

[@SørenJensen-f5q]

I had a Benchmark DAC1 HDR, and I was dissatisfied with the DAC's built in volume control. Doing volume control internally in JRiver sounded far better. Now I understand it probably had nothing to do with the HDR volume control. It was caused by reduced digital input level to the DAC due to Jriver volume control being operated below full scale.

[@aidmade85]

Thanks🙏How is this intersample clipping handled by real NOS DACs like Holo MAY?

[@klxz79]

AppleTV has the sound check in Apple Music which matches sound levels across songs. Maybe that acts as a -3DB adjustment?

[@Proper49er]

Good question. I wonder if miniDSP can let us know, not ideal purple light communication with the user.

[@Audioholics]

No, it's just a normalizer to keep all songs at around the same level.

[@scanspeak00]

I love this geeky stuff. More please.

[@jedi-mic]

The mini DSP when did that change over I thinking about buying a second hand one what would be the code between the new one and the old one you know thanks ! Could you achieve this by say a coaxial digital output from a CD player, or a DD converter, the connecting lead if you reduce its output by 3 db would achieve the same thing and should sound good! any leads out there will have to make one? You could achieve the old recording in a high sample rate say 96 by having a low pass filter if this is causing issues, just roll it off at 25 Hz at -24 DB you could use a passive low pass compressor from your sauce to your digital recording

[@ahlbergmagnus]

You can buy a second hand one since the fix is in the firmware and not hardware as he says in the video at 28:51.

[@paulpaulzadeh6172]

USB port is not galvanic isolated in their DAC !

[@AlexandreLollini]

But if you have access to audio files, with the audible problem, you can manipulate them to solve the issue. On some tracks it is so obvious.

[@scottivlow9962]

Gene I noticed the video from 5 years ago got removed from your channel relating to upfiring speakers for Dolby Atmos being called snake oil. It was on my recommendation list but when I searched for it not even on results page.

[@mccririck01]

Why does it become more of a problem with oversampling?

[@justincgs]

I bet the majority consumer DAC/DSP audio implementations suffer from this and I bet the majority of studio and mastering "pro audio engineers" of the modern age suffer this issue in their chain.

[@Artcore103]

Yes, they might, but it depends. Mine does not, because if you're into PC+DAC audio chains and like to research things, then there's much to learn and implement. This is about managing volume levels in various parts of the chain. My power amp has no volume control, so I only use digital volume which is actually ideal when you're running 24/32 bit dacs (and corresponding settings on the computer), you have PLENTY of dynamic headroom to significantly attenuate the signal without ANY meaningful impact or loss. So in windows there's a few things you have to keep in mind to avoid clipping, both the intersample clipping, and the windows audio stack's built in limiter which triggers before 0db level. I run eqapo for a complex parametric EQ but also use it's preamp gain filter to set a global/final (with EQ factored in) -3db output setting. Then additionally on foobar I use replaygain to do another -3db to all music. With other applications you can use their volume control as well and shave off a few db in them as well. With eqapo you can also disable all windows apos (the windows audio mixer stuff it's doing in the background) and thereby bypass ANY windows manipulation of the signal, and bypass that limiter entirely, if you wanted to. It also prevents you from being able to use any of the audio "enhancements", but I find the loudness enhancement quite useful for casual KZbin viewing and stuff... It's misnamed, it's actually a compressor, and a good one... It just levels the volume of everything, so it's less dynamic, keeps dialogue very audible and nothing gets too loud. So I use that but not for music or movies (unless it's casual and not audio-critical watching). I never use the windows resampler either, if I resample at all, it's with a good foobar vst plugin, otherwise I always avoid resampling in windows and set the output to match the source material. These things prevent intersample peaks (especially when resampling) and hitting the windows volume limiter, which wouldn't typically be audible to most, under most circumstances, but will technically harm the signal if you run into it. Can't do anything about the issues baked into certain recordings, but you can avoid adding your own issues on top. If the recording has 0db samples then it will clip between some of those samples - that actually can be solved via the volume/preamp gain management outlined above, but that's only when clipping isn't baked into the music due to the subsequent mixing/mastering steps... If there's clipping already within the music you can't fix that. But if there's no clipping baked in, only 0db samples that WOULD result in clipping during playback with full scale digital output, that can be avoided by bringing that first stage of gain down, so the +3 to +6db intersample peaks are actually kept below 0db, because you're starting at -6 or whatever, leaving room for the reconstructed peaks to not clip. There are good scientific articles and testing on this that I have referenced for all of this.

[@joesmith4443]

@@justincgs Loudness Wars

[@justincgs]

@@joesmith4443 Exactly

[@joesmith4443]

@@justincgs Regardless of how great Benchmark specs are and how their DACs don’t clip and calibrated perfectly if the 99% of music production doesn’t care about clipping, distortions, drop outs and warped music files it’s moot to do these videos in many ways. I’ve told Amir from ASR this during one of his 1,000,000 DAC recommendations that’s it more beneficial to also measure popular music production equipment and music files and how they measure after playback on great spec’d DACs then give them reviews. There’s no “subjectivity” in recording bad!

[@shodan6401]

I am a dummy. I know nothing about this stuff. But I'm asking honestly, is "Headroom" even an appropriate term within the context of the Digital Domain? It's data, not amplitude. And Interpolation is basically an informed guess. This is gonna take a while for me to understand...

[@miltono.1152]

I have to disagree on clipping isn't caused by incorrectly produced recordings, this issue can be solved beforehand so the end user won't have to apply -X dB in pre amp. No CDs came with a sticker "for better experience set pre amp -X dBs". It's like selling you something defective without telling you

[@paulpaulzadeh6172]

Amir test it , it was not so good as he claims !

[@owenjbrady]

is any of this actually auditable tho

[@paulpaulzadeh6172]

20 v per div ?? so high from DAC?! AP shows 20Vrms , something is fishy

[@jeroenk3570]

He's using BNC test leads instead of 10:1 probes and didn't set the input menu settings accordingly . I do that all the time if I just want to see the picture and don't feel like setting the input menu correctly, it's not a problem.